Jozu Hub + Agent Guard

Jozu is an on-premises security and governance platform for AI/ML workloads running on Kubernetes. It sits between training pipelines and inference serving, adding security controls that tools like KubeFlow and KServe do not natively provide. The platform has two primary components: - Jozu Hub: A model registry that packages AI models, agents, and MCP servers as ModelKits using the CNCF KitOps project and the CNCF ModelPack specification. It performs automated vulnerability and license scanning of model artifacts, cryptographically signs them, generates SBOMs, and enforces deployment governance policies before models reach inference serving. - Jozu Agent Guard: A runtime policy enforcement component for AI agents and MCP servers. It provides tool-level access control, guardrail policies, and human-in-the-loop enforcement within contained execution environments. Jozu installs behind the customer's firewall (including air-gapped environments) via a Helm chart. It is designed to integrate with existing Kubernetes AI stacks without replacing them. Key security capabilities include: - Automated multi-vector security scanning of model artifacts and dependencies - Cryptographic signing and SHA-based tamper-proof attestation - SBOM generation for supply chain security - Immutable model storage - Full audit trails for EU AI Act, ISO 42001, and NIST AI RMF compliance - Deployment governance and approval workflows - Runtime policy enforcement (ToolPolicy, GuardrailPolicy) via Agent Guard - OCI layer-deduplication for optimized model delivery (Rapid Inference Containers) - Hugging Face model import with security scanning Jozu does not replace experiment tracking or pipeline orchestration tools. It fills the gap between model development and production deployment by adding signing, scanning, and governance controls.